Principes d'invariance par moyennisation logarithmique pour les processus de Markov. (Invariance principles with logarithmic averaging for Markov processes) (Q1872250)

The author studies the almost sure central limit theorem for functionals of a recurrent Markov chain \((X_t)_{t\in T}\) with index set either \(T= \mathbb{N}_0\) or \(T= \mathbb{R}_+\) and with general state space \(S\). More specifically, additive martingale functionals are considered, i.e. processes \((M_t)_{t\in T}\) satisfying \(M_{t+s}= M_t+ M_s\circ \theta_t\) as well as \(E_x(M_t)= 0\), where \(\theta_t\) denotes the shift operator on the canonical probability space \(S^T\). In the present paper two main results are established, namely (i) an almost sure version of the functional central limit theorem for \((M_t)_{t\in T}\) and (ii) a central limit theorem and a law of the iterated logarithm extension of the strong law implicit in the almost sure CLT. The latter result only holds in the case of a positive recurrent Markov chain.